Published on

IJERA (International journal of Engineering Research and Applications) is International online, ... peer reviewed journal. For more detail or submit your article, please visit www.ijera.com

  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide


  1. 1. Misbah Siddiqui, Saba Khalid,Prof Rizwan Beg/ International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 5, September- October 2012, pp.008-013THE ANALYSIS OF VARIOUS TCP SUB-VERSIONS AND MECHANISM FOR CONGESTION AVOIDENCEMisbah Siddiquie Saba Khalid Prof Rizwan BegStudent Btech. CSE Lecturer CSE dept. Professor and HOD(CSE/IT)Integral University Integral University Integral UniversityLucknow, India Lucknow, India Lucknow, IndiaABSTRACT TCP, the most widely used protocol onInternet, has a major problem in that its The basic approach is to model the numbercongestion control does not allow flows to obtain of STAs with ACKs and data packets in their MACfull bandwidth on fast-long distance links. A queues as an embedded Discrete Time Markov ChainPerformance analysis of TCP-controlled long file (DTMC) and aggregate TCP throughput in thetransfers in a WLAN in infrastructure mode also Frame work of Renewal Reward theory(RRT) givenwith Comparison and Analysis of Congestion in[1] and [4].Window for HS-TCP, Full-TCP and TCP-Linux The concepts of the mechanism of TCPin Long Term Evolution System Model is congestion control depending dynamically onavailable in the literature with one of the main organize the size of window according to the networkassumptions being equal window size for all TCP path state. TCP has critical design and the congestionconnections. In this paper, we extend the analysis control permits to TCP to adjust the network end-to-to TCP-controlled long file uploads and end connection by control the data rate. According todownloads with different TCP windows. Our the available bandwidth. However, TCP give poorapproach is based on the semiMarkov process performance in high bandwidth channels due to theconsidered in [1] and [2], but with arbitrary slow response with large congestion window [1]. Ifwindow sizes. We present simulation results to TCP sender not receives Acknowledgment (ACK) forshow the accuracy of the analytical model. TCP receiver after transmission segment, the connection goes to timeout where the first phase ofKEYWORDS:-WLAN, ACCESS POINTS, start transmission called slow-start. In slowstart phaseINFRASTRUCTURE MODE, UPLOADING AND the packet sending and the window increasingDOWNLOADING, TCP WINDOW FULLTCP, HSTCP, exponentially in parallel with increasing the roundTCP-LINUX, CWND, LTE, NS-2. trip time(RTT).The window increment not continues without limits, because that will cause packet losses when the packets injected in network connection1. INTRODUCTION larger than the available space in network bottleneck In the previous papers a performance and receiver buffers. To overcome this defect of TCPanalysis of TCP-controlled long file transfers in a slow start, the researchers have proposed manyWLAN in infrastructure mode along with comparison improved methods. Fig. 1 shows the phases ofand analysis of congestion window for HS-TCP, standard cwnd. The basic congestion controlFull-TCP and TCP-Linux in Long Term Evolution mechanism supported by TCP include followingSystem Model are given, we have propose both phases, Slow-Start, Congestion Avoidance, Fastperformance analysis in terms of Aggregate Retransmit, and Fast Recovery.throughput and congestion window. This paper isconcerned with infrastructure mode WLANs that use TCP congestion control includes two main phases,the IEEE 802.11 DCF mechanism. We are interested slow start phase and congestion avoidance phase, andin analytical models for evaluating the performance these two phases used by TCPO sender to adjust theof TCP controlled simultaneous uploads and amount of data flow through network. TCP senderdownloads where each connection has arbitrary TCP starts by sending packets inform of segment andwindow size. A detailed analysis of the aggregate waiting to receive the ACK to proving the sentTCP download throughput in a WLAN for a single packets if it’s received or lost. If segmentrate Access Point (AP) is given in [1] and[3] where it acknowledged, the sender sends two segmentsis assumed that all TCP window sizes are equal. instead of one. After other ACK, the increment inSimilarly, aggregate TCP throughput is evaluated for congestion window duplicates to four. Actually, therethe multi rate case are in [4] and [5]. However, these is no exponential incrementing congestion windowworks also consider only download or upload with due to sometimes TCP sender receives delayed ACKconstant windows. and the sender transmits two segments every ACK. 8|Page
  2. 2. Misbah Siddiqui, Saba Khalid,Prof Rizwan Beg/ International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 5, September- October 2012, pp.008-013The possibility of congestion to occur in network TCP’s experimented over a traffic model of LTEconnections, when the network resources supply a network.large amount of data exceed the available bandwidthand these problem represents a big challenge inmodern networks due to its cause up normaltransferring in packets, like delayed packets and lostpackets, where that will force the sender toresendthese packets and the performance of networkwillsuffering from degradation [2]. Each transmissionperiod starts with slow start phase andthe segmenttransfer from TCP sender to its peer untilthecongestion window become equal to the presentthreshold(ssthresh) of the network pipeline. At thispoint, the network enters to congestion avoidanceperiod and the window starts increasing linearly. 2. RELATED WORK The analytical work in this area has considered saturated and unidirectional traffic, i.e, either uplink or downlink; see, for example, [5], [6] and [7]. All the above papers assume that all the STAs are saturated, in other words they have packets to send to the AP at all time. In contrast, we consider TCP controlled transfers, where the "saturated nodes assumption "does not apply. All the related work that we are aware of assumes homogeneous TCP connections in the sense that the maximum window size is the same for all connections. [1] and [8] propose a model for a single rate AP-STA WLAN, assuming the same maximum size of TCP window for all TCP connections. An extension of this work in [2] considers two rates of association with long file uploads from STAs to a local server; the multi rate case is considered in [3]. [9] and [10] present analysis of TCP-controlled uploads and downloads with UDPtraffic in a single cell infrastructure WLAN. They assume equal TCP maximum window size for In other words, the window enlarges by one all connections, that TCP receivers use undelayedsegment per RTT. [3]. There are two other phases ACKs, and show that the totalwhich support some other TCP variant these are fast TCP throughput is independent of the number ofretransmit and fast recoveries . Fast retransmit STAs in the system. Also, upload and downloadmechanism allows to TCP sender to retransmit a transfers obtain equal shares of the total throughput.segment after specific timeout period and no need to The letter [11]gives an average value analysis of TCPwait ACK from receiver. performance with upload and download traffic. First, the authors provide an expression for the average This mechanism considers the packet lost if number of active TCP stations. In [12], a finite bufferACK delay [4].In other hand, when fast retransmit AP with TCP traffic in both upload and downloaddetects three duplicate ACKs, the fast recovery direction is analysed with delayed and undelayedprocess starts from congestion avoidance region and ACK cases.[13] provides an analysis for a givenuse ACKs in the pipe to transmitting packets. This number of STAs and maximum TCP receive windowarticle provides comparison and investigation to the size by using the well known p-persistent model andbehavior of congestion window for three TCP in system model proposed in [1]&[7]. However, [13]versions; HSTCP, FullTCP, and TCP-Linux with considers only download traffic or upload traffic, notanalyzing the congestion window clocking and both together. HTTP traffic is analyzed in [14]. Atiming when these three queueing model is proposed to compute the mean session delay in the presence of short-lived TCP flows. The impact of TCP maximum congestion 9|Page
  3. 3. Misbah Siddiqui, Saba Khalid,Prof Rizwan Beg/ International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 5, September- October 2012, pp.008-013window size on this delay is studied. The analysisalso wired and wireless systems, although TCP was notextended to consider the delayed ACK technique. originally designed for real time applications and not for wireless networks then, we need to develop a new Our contribution: We provide a simple TCP mechanisms, or at least chooseapproach to model the aggregate throughput of long- a suitable TCP variant for each new network, to belived TCP downloads and uploads also with more efficient and more reliability with this network.Comparison and Analysis of Congestion Window forHS-TCP, Full-TCP and TCP-Linux in Long Term 3. PROPOSED TOPOLOGY ANDEvolution System Model given in [1]and[2]. NETWORK SIMULATION we used NS-2 network simulator to2.1. BRIEF DESCRIPTION OF TCPS AND LTE modulate the proposed topology of LTE and also theSYSTEMS experiments to monitor the behaviour of TCP HSTCP introduce a new mechanism based versions used in these article performed according toon enhancement the time of loss recovery of classic NS-2, where NS-2 is a better simulation deals withTCP by variation the standard algorithm of TCP’s TCPs protocols. NS-2 not just a simulator, but it’s aAdditive discrete event aimed to support the research and Increase Multiplicative Decrease (AIMD). studies that deals with communications and networksThe developed mechanism of HSTCP works analysis. Also, NS-2 provides environments toeffectively with large congestion windows. That simulate and modeling multicast protocols, networksmean, when the congestion window less than the traffics, handovers, and other networks resources andthreshold, HSTC Puses AIMD algorithm to control conditions for wireless and wired channels [12]. Incongestion in network connections, but if congestion our research, we used NS-2 version 2.32 , and thiswindow high its use the algorithm of high speed version installed over WindowsAIMD [5]. Actually, designing of XP or using Cygwin, where Cygwin HSTCP is depending on the response provides a Linux-like environment under Windows,variation in environments of the low congestion because NS-2 already supported a Linux operatingoccurring in bottleneck and to the response activation system only, then we need to get a virtualof standard TCP in environments of packet loss rates environment[6].FullTCP can only apply with the congestioncontrol of TCPReno. However, FullTCP must beprovided with fully algorithms of congestion controlsuch as Tahoe, Vegas, and Sack [7]. In TCP-Linux,it’s introduced three main characteristics todeveloping the performance of TCP-Linux.These three elements are [8]:• Standard interface for congestion controlalgorithms.• Redesigned loss detection module.• New event queue scheduler that increases the testspeed.As we mentioned before, this study and theexperimental test depended on analysis the behaviourof three TCPs over LTE network topology. LTE isthe 3GPP specification for the fourth generation ofmobile networks and referred to as Evolved UMTSTerrestrial Radio Access (E-UTRA). LTE systemsaim to provide a step forward in wireless and mobile SIMULATION PARAMETERSsystems by providing high speed data transmission to Parameter Valueusers by providing low latency links an improved TCP protocolsspectral efficiency. Comparing with Wideband Code HSTCP, FullTCP, TCPLinuxDivision Multiple Access Propagation Delay of all links 2 msec(WCDMA) and HSPA, LTE provides higher rate of Bandwidth of a GW link 10 Mbpsdata with downlink reaches more than 100Mbps Bandwidth of Server 100 Mbpsuplink of 50 Mbps. Infact, LTE systems released Bandwidth of UE link 1Mbpswide achievements in telecommunication networks Packet Size 1500 Bytesby providing lower user costs than other systems, Window size 100 Kbytesbetter spectrum efficiency, and a very small latency Simulation Time 50 sec[9, 10].LTE systems is under developments now, and the As shown in Fig. 3, we proposed six nodesTCP protocol is the most widely used protocol for elements. Two nodes as mobile nodes (UEs), two 10 | P a g e
  4. 4. Misbah Siddiqui, Saba Khalid,Prof Rizwan Beg/ International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 5, September- October 2012, pp.008-013base stations (eNBs), one gateway router, and oneFTP server. The proposed LTE model is assumed a /* FULLTCP agent */bottleneck connection with bandwidth of 100Mbps set tcp [new Agent/TCP/FullTCP]and delay of 2 msec. The parameters of modeling and set tcpsink [new Agent/TCP/FullTCP]simulation presented in Table 1, where all links areset to unified propagation delay of 2 msec. Themaximum packet size of TCP was set to 1500 Bytes, 4. RESULTS ANALYSIS ANDwith minimum window size of 100 Kbytes. NS-2 DISCUSSIONSrepresentation of links depend on bandwidth of the The results obtained from experimentslink, type of link (duplex or simplex) and the divided into four parts, three parts to analyze thepropagation delay of packets over this link. In LTE slow-start and congestion avoidance phases of eachnetworks simulation, we have many bandwidths and TCP, and the last compares the full period of cwnd oftwo types of links but only one delay for all links. three TCPs in one experiment.As we mentioned before, NS-2 provides the agents of As shown in Fig. 4, the slow-start phase ofmost of TCP’s including the three TCP’s studied TCP-Linux shows that it reaches to congestion withinhere. TCP-Linux not provided directly in NS-2.32, 1.25 sec, and the maximum window reach to 100but it need to install it by using a compatible patch Kbyte. It have a congestion avoidance after 1.5 secand process some modification in NS-2 and then it start a new slow-start phase with newenvironments. In TCP-Linux, TCP agent name ssthresh at 1.5 sec to a new congestion point at 50change from (Agent/TCP/Sack1) to Kbytes. Thats mean ssthresh=cwnd/2, since TCP-(Agent/TCP/Linux), and the TCP destination set to Linux implementation evaluates the initial ssthreshSack1. from the cwnd size of the previous TCP connection. However, to implementing TCP-Linux inTCL script (TCL is the language used inprogramming inside NS-2), we must choose one oftwo available agents form, either using the agent inthe form of: (Agent/TCPSink/Sack1) or using theagent in form: (Agent/TCPSink/DelAck). As shownin script below: /* TCP-Linux agent */set tcp [new Agent/TCP/Linux]set tcpsink [new Agent/TCPSink] In HSTCP, the structure of agents differsfrom that used in TCP-Linux. The implementation ofHSTCP is controlled by the manner to adjusting thecongestion control. This TCP variant depends onmodifying the response function and only determinesthe effect when the congestion window is high. InNS-2, implementing of HSTCP to test the congestionwindow behavior or for any other application alsoneeds to use TCP source and destination withdifferent agents. When HSTCP is applied in NS-2, Fig. 4 Slow start for TCP Linuxmust use the form shown below:/* HSTCP agent */set tcp [new Agent/TCP/Sack1]set tcpsink [new Agent/TCPSink/Sack1] The implementation of FullTCP in NS-2 isstill under development because of its adding newlyto the TCP suite of the agents which supported byNS-2 configuration. FullTCP agent is differ from theother agents used under NS-2 due to Full TCP agentis classified as bidirectional protocol. That means,FullTCP is a two way TCP where its providebidirectional data transmission. In addition, FullTCPsupport a new data structure, where it uses the bytesinstead of packets in representing the sequencenumbers. FullTCP can implementing in TCL script inthe form shown below: 11 | P a g e
  5. 5. Misbah Siddiqui, Saba Khalid,Prof Rizwan Beg/ International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 5, September- October 2012, pp.008-013 The three TCPs can provide a good and regular cwnd, inspite of the differences is slow-start phase behavior. As shown in Fig. 7, the maximum slow- start founded in TCP-Linux where it exceeded 1 sec, with ssthresh of 50 Kbytes with small congestion avoidance. But cwnd of FullTCP performed a better ssthresh of 60Kbytes. A zooming graph of one complete period of cwnd of three TCPs, shown in Fig. 8, where the maximum congestion point of network got with FullTCP, in other side, the minimum point appeared with HSTCP. The maximum slow-start phase occurred with TCP-Linux with maximum congestion avoidance of 1 Kbyte, where, lager congestion avoidance increment will improve performance.Fig. 5 Slow start for Full TCP In this Fig. 5, the slow-start phase of FullTCPneeds only to 0.5 sec to reach window of 100 Kbytes,and no clearance congestion avoidance here, but wecan note some fast recovery with some packets lost.The new slow-start phase starts after 1 sec to reach anew congestion point at 60 Kbytes. Slow-start ofFullTCP, starts unstable, but after first 1 sec did welland provided a stable behaviour of cwnd.From Fig. 6 we illustrate the slow-start of HSTCP,where it reached within 1 sec, to 80 Kbytes asssthresh, without congestion avoidance. So it needslong slow-start time for new session, reached to 10seconds. Whatever, it provided a regular cwnd but it Fig. 7 cwnd of HSTCP, TCP-Linux, and FullTCPhad a small window compared with other twovariants. Fig. 8 Enlarged plot of Slow-Start and Congestion Avoidance Fig. 6 Slow-Start phase of HSTCP When cwnd can reach maximum value, the losses in packets may occur and performance 12 | P a g e
  6. 6. Misbah Siddiqui, Saba Khalid,Prof Rizwan Beg/ International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 5, September- October 2012, pp.008-013decrease. In the plot here at Fig. 8, the cwnd can be Fi Hot Spot Networks," Networking 2004,larger than the bandwidth delay product, we only get LNCS 3042, pages 626- 637,200460 Kbytes in FullTCP session as a throughput. The [6] R. Bruno, M. Conti, and E. Gregori,losses occur when the window size becomes larger "Throughput analysis and measurements inthan the point of congestion window of the network, IEEE 802.11 WLANs with TCP and UDPor we can say the packets lost because the network traffic flows," IEEE Transactions on Mobilehas not enough capacity for storing all the sent Computing, vol.7, no.2, pp. 171-186, Feb.packets, so, if the window size is too large, some 2008.packets are dropped and performance decreases, [7] R. Bruno , M. Conti, and E. Gregori, "Ansince when a loss occur, the cwnd is divide by two. accurate closed-form formula for the throughput of long-lived TCP connections in3. CONCLUSIONS IEEE 802.11 WLANs; Computer Networks, This paper provides a mechanism to the vol. 52, pp. 199-212,2008behavior of congestion window for different TCP [8] R Bruno, M Conti, and E Gregori, "Averageversions, HSTCP, FullTCP, and TCP-Linux, over a value analysis of 802.11 WLANs withmodel of LTE network using NS-2 network persistent TCP flows," IEEEsimulator. The aim of this work was to establish a Communication Letters, vol.13, noA, Apr.bound on the performance of cwnd of these TCPs in 2009.high bandwidth networks also with Performance [9] O. Bhardwaj, G. V. V. Sharma, M. Panda,analysis of TCP-controlled long file transfers. The and A. Kumar, "Modeling finite bufferinvestigation has been carried out by modelling all effects on tcp traffic over an ieee 802.11links and traffics of LTE networks with standard infrastructure wlan," Computer Networks,parameters, such as bandwidth and delay, to monitor vol. 53, pages. 28552869, Nov 2009the cwnd, slow-start, and congestion avoidance of [10] J. Yu, S. Choi, "Modeling and analysis ofthese variants. Full TCP provided a well performance TCP dynamics over IEEE 802.11 WLAN,"and a big widow size compared with other two Proceedings o/ the IFlP WONS" pp. 154-variants, so HSTCP had a longest slow-start period 161 Obergurgl, Austria, 2007with a congestion level near to that produced by [11] K. Fall, S. Floyd, and T. Henderson, “NsTCP-Linux. Generally, the implementation of three simulator tests for reno fulltcp,” U.S.TCPs shows the differences in behaviour of Department of Energy,1997.congestion window phases in spite of it is a little but [12] D. X. Wei and P. Cao, “NS-2 TCP-Linux: Anwhen used over high bandwidth or any other high NS-2 TCP Implementation with Congestionspeed networks, we can see a large rate of packets Control Algorithms from Linux,” Workshoplosses, because of incompatibility between the on NS-2, October 2006.congestion window parameters and the link capacity. [13] T. Mahmoodi,“ Transport Layer Performance Enhancements over WirelessREFERENCES Networks,” University of London, August [1] Ghassan A. Abed Mahamod Ismail 2009. Kasmiran Jumari “A Comparison and [14] Y. Tan,“Active Queue Management for LTE Analysis of Congestion Window for HS- Uplink in eNodeB,” Master Thesis, Helsinki TCP, Full-TCP,and TCP-Linux in Long University of Technonlogy, February 2009. Term Evolution System Model” in IEEE [15] A. Racz, A. Temesvary, and N. Reider, conference on open system(ICOS2011). “Handover Performance in 3GPP Long [2] Pradeepa BK and Joy Kuri ”Aggregate AP Term Evolution (LTE) Systems,” Mobile Throughputs for Long File Transfers in a and Wireless Communications Summit, WLAN controlled by Inhomogeneous TCP” 2007. 16th IST, pp. 1–5, July 2007. Connections in proc. of IEEE 2011. [16] The Network Simulator - NS-2. In: [3] A. Kumar, E. Altman, D. Miorandi, and M. Information Sciences Institute, University of Goyal, "New insights from a fixed point Southern California, USA, 2006. analysis of single cell IEEE 802.11 WLANs," IEEE ACM Transactions on Networking, Vol. 15, No 3, pp. 588-601, June 2007. [4] F. Cali, M. Conti, and E. Gregori, "Dynamic Tuning of the IEEE 802.11 protocol to achieve a Theoretical Throughput Limit; IEEElACM Transaction On Networking, vol. 8 no. 6 pp. 785-799, Dec 2000 [5] R. Bruno,M . Conti,a nd E. Gregori," Throughput Analysis of TCP clients in Wi- 13 | P a g e